I have two continuous variables: $X1$ and $X2$, both have a positive correlation on the dependent variable $y$ (continuous). I found that the interaction term $(X1*X2)$ is statistically significant for $p\lt(0.00)$ when added to the model. The final model is the following:
Where $a$, $b$, and $c$ are the regression coefficients. However, I am wondering why the regression coefficient for $c$ of the interaction term is negative? What does it mean?
If I only add the interaction term $(X1*X2)$ to the model why is the coefficient positive when if I add it with $(X1)$ and $(X2)$ the regression coefficient becomes negative? It doesn't make sense!